[Feature request] Bulk polling API for perf ring buffer

Perf ring buffer (or perf buffer) created with BPF_PERF_OUTPUT(). BPF runtime can submit events to perf buffer via perf_submit() API. Userspace call PerfBuffer::poll() to invoke callbacks assigned during the initialization of the perf buffer.

IIUC, the perf buffer polling will repetitively invoke the callback on each and every event in the perf buffer. That can be wasteful when there is a lot of events flowing through the perf buffer.

Instead, I am wondering if a bulk polling API would be more efficient.

Start by reading the BPF_PERF_OUTPUT() and perf_submit() flow, then trace the userspace PerfBuffer::poll() entry point and its callback behavior. The payload names no files or tests, so locate the relevant implementation and coverage before defining the bulk API. Done means the API scope, callback semantics, and validation approach are agreed and implemented.

Written by the indexing model from the issue text.